Merge branch 'master' into akihironitta/bench

.github/actions/setup/action.yml

@@ -49,12 +49,13 @@ runs:
     - name: Install pyg-lib  # pyg-lib is currently only available on Linux.
       if: ${{ inputs.torch-version != 'nightly' && runner.os == 'Linux' }}
       run: |
-        pip install pyg-lib -f https://data.pyg.org/whl/nightly/torch-${{ inputs.torch-version }}+${{ inputs.cuda-version }}.html
+        pip install --no-index pyg-lib -f https://data.pyg.org/whl/nightly/torch-${{ inputs.torch-version }}+${{ inputs.cuda-version }}.html
       shell: bash
 
     - name: Install extension packages
       if: ${{ inputs.full_install == 'true' && inputs.torch-version != 'nightly' }}
       run: |
         pip install torchvision==${{ inputs.torchvision-version }} --extra-index-url https://download.pytorch.org/whl/${{ inputs.cuda-version }}
-        pip install torch-scatter torch-sparse torch-cluster torch-spline-conv -f https://data.pyg.org/whl/torch-${{ inputs.torch-version }}+${{ inputs.cuda-version }}.html
+        pip install scipy
+        pip install --no-index torch-scatter torch-sparse torch-cluster torch-spline-conv -f https://data.pyg.org/whl/torch-${{ inputs.torch-version }}+${{ inputs.cuda-version }}.html
       shell: bash

.github/workflows/full_testing.yml

@@ -16,8 +16,12 @@ jobs:
       matrix:
         os: [ubuntu-latest, windows-latest]
         python-version: ['3.8', '3.10']
-        torch-version: [1.13.0, 2.0.0, nightly]
+        torch-version: [1.11.0, 1.12.0, 1.13.0, 2.0.0, nightly]
         include:
+          - torch-version: 1.11.0
+            torchvision-version: 0.12.0
+          - torch-version: 1.12.0
+            torchvision-version: 0.13.0
           - torch-version: 1.13.0
             torchvision-version: 0.14.0
           - torch-version: 2.0.0

.github/workflows/prev_testing.yml

@@ -1,4 +1,4 @@
-name: Testing PyTorch 1.13
+name: Testing previous PyTorch versions
 
 on:  # yamllint disable-line rule:truthy
   push:
@@ -11,6 +11,18 @@ jobs:
   prev_pytest:
     runs-on: ubuntu-latest
 
+    strategy:
+      fail-fast: false
+      matrix:
+        torch-version: [1.11.0, 1.12.0, 1.13.0]
+        include:
+          - torch-version: 1.11.0
+            torchvision-version: 0.12.0
+          - torch-version: 1.12.0
+            torchvision-version: 0.13.0
+          - torch-version: 1.13.0
+            torchvision-version: 0.14.0
+
     steps:
       - name: Checkout repository
         uses: actions/checkout@v3
@@ -36,8 +48,8 @@ jobs:
         if: steps.changed-files-specific.outputs.only_changed != 'true'
         uses: ./.github/actions/setup
         with:
-          torch-version: 1.13.0
-          torchvision-version: 0.14.0
+          torch-version: ${{ matrix.torch-version }}
+          torchvision-version: ${{ matrix.torchvision-version }}
 
       - name: Install main package
         if: steps.changed-files-specific.outputs.only_changed != 'true'

CHANGELOG.md

@@ -7,31 +7,43 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 ### Added
 
+- Added back support for PyTorch >= 1.11.0 ([#7656](https://github.com/pyg-team/pytorch_geometric/pull/7656))
+- Added `Data.sort()` and `HeteroData.sort()` functionalities ([#7649](https://github.com/pyg-team/pytorch_geometric/pull/7649))
+- Added `torch.nested_tensor` support in `Data` and `Batch` ([#7643](https://github.com/pyg-team/pytorch_geometric/pull/7643), [#7647](https://github.com/pyg-team/pytorch_geometric/pull/7647))
+- Added `interval` argument to `Cartesian` and `LocalCartesian` transformations ([#7533](https://github.com/pyg-team/pytorch_geometric/pull/7533), [#7614](https://github.com/pyg-team/pytorch_geometric/pull/7614))
+- Added a `LightGCN` example on the `AmazonBook` dataset ([7603](https://github.com/pyg-team/pytorch_geometric/pull/7603))
+- Added a tutorial on hierarchical neighborhood sampling ([#7594](https://github.com/pyg-team/pytorch_geometric/pull/7594))
+- Enabled different attention modes in `HypergraphConv` via the `attention_mode` argument ([#7601](https://github.com/pyg-team/pytorch_geometric/pull/7601))
+- Added the `FilterEdges` graph coarsening operator ([#7361](https://github.com/pyg-team/pytorch_geometric/pull/7361))
+- Added the `DirGNN` model for learning on directed graphs ([#7458](https://github.com/pyg-team/pytorch_geometric/pull/7458))
+- Allow GPU tensors as input to `NodeLoader` and `LinkLoader` ([#7572](https://github.com/pyg-team/pytorch_geometric/pull/7572))
+- Added an `embedding_device` option to allow for GPU inference in `BasicGNN` ([#7548](https://github.com/pyg-team/pytorch_geometric/pull/7548))
+- Added `Performer` to `GPSConv` and remove `attn_dropout` argument from `GPSConv` ([#7465](https://github.com/pyg-team/pytorch_geometric/pull/7465))
 - Enabled `LinkNeighborLoader` to return number of sampled nodes and edges per hop ([#7516](https://github.com/pyg-team/pytorch_geometric/pull/7516))
 - Added the `HM` personalized fashion recommendation dataset ([#7515](https://github.com/pyg-team/pytorch_geometric/pull/7515))
-- Added the `GraphMixer` model ([#7501](https://github.com/pyg-team/pytorch_geometric/pull/7501))
+- Added the `GraphMixer` model ([#7501](https://github.com/pyg-team/pytorch_geometric/pull/7501), [#7459](https://github.com/pyg-team/pytorch_geometric/pull/7459))
 - Added the `disable_dynamic_shape` experimental flag ([#7246](https://github.com/pyg-team/pytorch_geometric/pull/7246), [#7534](https://github.com/pyg-team/pytorch_geometric/pull/7534))
 - Added the option to override `use_segmm` selection in `HeteroLinear` ([#7474](https://github.com/pyg-team/pytorch_geometric/pull/7474))
 - Added the `MovieLens-1M` heterogeneous dataset ([#7479](https://github.com/pyg-team/pytorch_geometric/pull/7479))
 - Added a CPU-based and GPU-based `map_index` implementation ([#7493](https://github.com/pyg-team/pytorch_geometric/pull/7493))
 - Added the `AmazonBook` heterogeneous dataset ([#7483](https://github.com/pyg-team/pytorch_geometric/pull/7483))
-- Added hierarichial heterogeneous GraphSAGE example on OGB-MAG ([#7425](https://github.com/pyg-team/pytorch_geometric/pull/7425))
-- Added the `torch_geometric.distributed` package ([#7451](https://github.com/pyg-team/pytorch_geometric/pull/7451), [#7452](https://github.com/pyg-team/pytorch_geometric/pull/7452))
+- Added hierarchical heterogeneous GraphSAGE example on OGB-MAG ([#7425](https://github.com/pyg-team/pytorch_geometric/pull/7425))
+- Added the `torch_geometric.distributed` package ([#7451](https://github.com/pyg-team/pytorch_geometric/pull/7451), [#7452](https://github.com/pyg-team/pytorch_geometric/pull/7452)), [#7482](https://github.com/pyg-team/pytorch_geometric/pull/7482), [#7502](https://github.com/pyg-team/pytorch_geometric/pull/7502), [#7628](https://github.com/pyg-team/pytorch_geometric/pull/7628))
 - Added the `GDELTLite` dataset ([#7442](https://github.com/pyg-team/pytorch_geometric/pull/7442))
 - Added the `approx_knn` function for approximated nearest neighbor search ([#7421](https://github.com/pyg-team/pytorch_geometric/pull/7421))
 - Added the `IGMCDataset` ([#7441](https://github.com/pyg-team/pytorch_geometric/pull/7441))
 - Added a sparse `cross_entropy` implementation ([#7447](https://github.com/pyg-team/pytorch_geometric/pull/7447), [#7466](https://github.com/pyg-team/pytorch_geometric/pull/7466))
 - Added the `MovieLens-100K` heterogeneous dataset ([#7398](https://github.com/pyg-team/pytorch_geometric/pull/7398))
-- Added the `PMLP` model ([#7370](https://github.com/pyg-team/pytorch_geometric/pull/7370))
+- Added the `PMLP` model and an example ([#7370](https://github.com/pyg-team/pytorch_geometric/pull/7370), [#7543](https://github.com/pyg-team/pytorch_geometric/pull/7543))
 - Added padding capabilities to `HeteroData.to_homogeneous()` in case feature dimensionalities do not match ([#7374](https://github.com/pyg-team/pytorch_geometric/pull/7374))
 - Added an optional `batch_size` argument to `fps`, `knn`, `knn_graph`, `radius` and `radius_graph` ([#7368](https://github.com/pyg-team/pytorch_geometric/pull/7368))
 - Added `PrefetchLoader` capabilities ([#7376](https://github.com/pyg-team/pytorch_geometric/pull/7376), [#7378](https://github.com/pyg-team/pytorch_geometric/pull/7378), [#7383](https://github.com/pyg-team/pytorch_geometric/pull/7383))
-- Added an example for hierarichial sampling ([#7244](https://github.com/pyg-team/pytorch_geometric/pull/7244))
+- Added an example for hierarchical sampling ([#7244](https://github.com/pyg-team/pytorch_geometric/pull/7244))
 - Added Kùzu remote backend examples ([#7298](https://github.com/pyg-team/pytorch_geometric/pull/7298))
 - Fixed tracing of `add_self_loops` for a dynamic number of nodes ([#7330](https://github.com/pyg-team/pytorch_geometric/pull/7330))
 - Added an optional `add_pad_mask` argument to the `Pad` transform ([#7339](https://github.com/pyg-team/pytorch_geometric/pull/7339))
 - Added `keep_inter_cluster_edges` option to `ClusterData` to support inter-subgraph edge connections when doing graph partitioning ([#7326](https://github.com/pyg-team/pytorch_geometric/pull/7326))
-- Unify graph pooling framework ([#7308](https://github.com/pyg-team/pytorch_geometric/pull/7308))
+- Unify graph pooling framework ([#7308](https://github.com/pyg-team/pytorch_geometric/pull/7308), [#7625](https://github.com/pyg-team/pytorch_geometric/pull/7625))
 - Added support for tuples as keys in `ModuleDict`/`ParameterDict` ([#7294](https://github.com/pyg-team/pytorch_geometric/pull/7294))
 - Added `NodePropertySplit` transform for creating node-level splits using structural node properties ([#6894](https://github.com/pyg-team/pytorch_geometric/pull/6894))
 - Added an option to preserve directed graphs in `CitationFull` datasets ([#7275](https://github.com/pyg-team/pytorch_geometric/pull/7275))
@@ -59,6 +71,10 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 ### Changed
 
+- Changed `add_random_edge` to only add true negative edges ([#7654](https://github.com/pyg-team/pytorch_geometric/pull/7654))
+- Allowed the usage of `BasicGNN` models in `DeepGraphInfomax` ([#7648](https://github.com/pyg-team/pytorch_geometric/pull/7648))
+- Breaking Change: Made `Data.keys` a method rather than a property ([#7629](https://github.com/pyg-team/pytorch_geometric/pull/7629))
+- Added a `num_edges` parameter to the forward method of `HypergraphConv` ([#7560](https://github.com/pyg-team/pytorch_geometric/pull/7560))
 - Fixed `get_mesh_laplacian` for `normalization="sym"` ([#7544](https://github.com/pyg-team/pytorch_geometric/pull/7544))
 - Use `dim_size` to initialize output size of the `EquilibriumAggregation` layer ([#7530](https://github.com/pyg-team/pytorch_geometric/pull/7530))
 - Added a `max_num_elements` parameter to the forward method of `GraphMultisetTransformer`, `GRUAggregation`, `LSTMAggregation` and `SetTransformerAggregation` ([#7529](https://github.com/pyg-team/pytorch_geometric/pull/7529))

benchmark/training/training_benchmark.py

@@ -17,6 +17,7 @@
     test,
     write_to_csv,
 )
+from torch_geometric import compile
 from torch_geometric.loader import NeighborLoader
 from torch_geometric.nn import PNAConv
 from torch_geometric.profile import rename_profile_file, timeit, torch_profile
@@ -191,6 +192,10 @@ def run(args: argparse.ArgumentParser):
                             metadata=data.metadata() if hetero else None)
                         model = model.to(device)
                         model.train()
+
+                        if args.compile:
+                            model = compile(model, dynamic=True)
+
                         optimizer = torch.optim.Adam(model.parameters(),
                                                      lr=0.001)
 
@@ -330,6 +335,7 @@ def run(args: argparse.ArgumentParser):
     add('--export-chrome-trace', default=True, type=bool,
         help='Export chrome trace file. Works only with PyTorch profiler')
     add('--trim', action='store_true', help="Use `trim_to_layer` optimization")
+    add('--compile', action='store_true')
     args = argparser.parse_args()
 
     run(args)

docker/Dockerfile

@@ -1,7 +1,7 @@
 FROM ubuntu:18.04
 
 # metainformation
-LABEL org.opencontainers.image.version = "2.1.0"
+LABEL org.opencontainers.image.version = "2.3.1"
 LABEL org.opencontainers.image.authors = "Matthias Fey"
 LABEL org.opencontainers.image.source = "https://github.com/pyg-team/pytorch_geometric"
 LABEL org.opencontainers.image.licenses = "MIT"

docs/source/advanced/cpu_affinity.rst

@@ -46,7 +46,7 @@ The recommended number of workers to start with lies between :obj:`[2, 4]`, and
         for batch in loader:
             pass
 
-It is generally adivisable to use :obj:`filter_per_worker=True` for any multi-process CPU workloads (:obj:`True` by default).
+It is generally advisable to use :obj:`filter_per_worker=True` for any multi-process CPU workloads (:obj:`True` by default).
 The workers then prepare each mini-batch: first by sampling the node indices using pre-defined a sampler, and secondly filtering node and edge features according to sampled nodes and edges.
 The filtering function selects node feature vectors from the complete input :class:`~torch_geometric.data.Data` tensor loaded into DRAM.
 When :attr:`filter_per_worker` is set to :attr:`True`, each worker's subprocess performs the filtering within it's CPU resource.